Hello, stranger?
Juliana Schroeder1, Donald Lyons2, & Nicholas Epley2
1University of California Berkeley Haas School of Business
2University of Chicago Booth School of Business
© 2021, American Psychological Association. This paper is not the copy of record and
may not exactly replicate the final, authoritative version of the article. Please do not
copy or cite without authors' permission. The final article will be available, upon
publication, via its DOI: 10.1037/xge0001118
Corresponding Author:
Juliana Schroeder
Author Note: We thank Astrid Martinez Cedillo, Sophie Ellis, Athanasios Ioannidis, Viktoriya
Koleva, Zaichen Li, Tegan Myhill, Rasheedat Olarinoye, Lila Stoddard, Maria Sanz Taberner,
and Tania Garcia Vite for assistance conducting experiments, Gillian Sandstrom and Rael
Dawtry for assistance recruiting research assistants, Emily Kasriel and staff at the British
Broadcasting Corporation for help securing permission to conduct experiments in UK train
stations, and the Booth School of Business and Haas School of Business for financial support.
Data and materials are available at:
https://osf.io/cd86s/?view_only=8b3ad7213eb6457399f6cf066a82d24a.
Abstract
Connecting with others makes people happier, but strangers in close proximity often
ignore each other. Prior research (Epley & Schroeder, 2014) suggested this social disconnection
stems from people misunderstanding how pleasant it would be to talk with strangers. Extending
these prior results, in a field experiment with London-area train commuters, those assigned to
talk with a stranger reported having a significantly more positive experience, and learning
significantly more, than those assigned to a solitude or control condition. Commuters also
expected a more positive experience if they talked to a stranger than in the solitude or control
conditions. A second experiment explored why commuters nevertheless avoid conversation even
when it is generally pleasant. Commuters predicted that trying to have a conversation would be
less pleasant than actually having one because they anticipated that others would be uninterested
in talking. These experiments clarify the precise aspects of social interaction that may be
misunderstood. People may avoid pleasant conversations with strangers because of miscalibrated
concerns about starting them.
Hello Stranger, 3
In an effort to encourage more conversation on the London Tube, a train rider in 2016
began distributing badges through the London Underground network emblazoned with the
question, “Tube Chat?” Soon afterward, a newspaper claimed that the campaign “provoked
horror among London commuters,”1 while a rival group launched a Twitter account called “Shut
Up Tube Chat.”2 Volunteers from the Shut Up Tube Chat campaign handed out anti-chat badges
that read, “Don’t even think about talking to me,” at the Liverpool Street tube station one Friday
morning.
Although Londoners have a reputation for being reserved (Boyce, 2012), prior research
suggests that their “horror” at imagining a conversation with strangers may be miscalibrated.
Human beings are inherently social, such that connecting with others—even with complete
strangers—reliably increases people’s moods (Dunn, Biesanz, Human, & Finn, 2007; Fleeson,
Malanos, & Achille, 2002; Gunaydin et al., 2020; McNiel & Fleeson, 2006; Sandstrom & Dunn,
2014; Zelenski, Santoro, & Whelan, 2012; Zelenski et al., 2013). Acting more extroverted in a
social setting also increases people’s positive mood, even among self-reported introverts, in both
laboratory settings (Fleeson et al., 2002; McNiel & Fleeson, 2006; McNiel, Lowman, & Fleeson,
2010; Zelenski et al., 2012), and weeks-long field interventions (Margolis & Lyubomirsky,
2020). In experiments on commuter trains, even people assigned to talk to a stranger do not
report experiencing “horrific” outcomes but instead report having a significantly more positive
commute than those assigned to keep to themselves in solitude or do whatever they normally do
(Epley & Schroeder, 2014).
Hello Stranger, 4
If connecting with a stranger can make a commute more pleasant than keeping to oneself,
then why do people often seem so reluctant to do it, like the Londoners wearing “don’t talk to
me” badges? One possibility is that people’s expectations about the outcomes of social
interactions are miscalibrated, such that they underestimate the positive consequences of
connecting with a stranger. In one laboratory experiment, participants expected to enjoy talking
to their own romantic partner more than talking to an opposite-sex stranger, but these
participants actually enjoyed talking to these two targets equally (Dunn et al., 2007). In a mix of
workshops and week-long field experiments (Sandstrom & Boothby, 2020), participants asked to
talk with strangers consistently underestimated how much they would like their partners, how
much they would enjoy their conversations, and how competent they would be in carrying on
their conversations. And in field experiments on trains and busses (Epley & Schroeder, 2014),
Chicago-area commuters expected that they would have a less pleasant commute if they
connected with a stranger in conversation than if they sat in solitude or had their typical
commute. In fact, participants randomly assigned to do one of these three activities actually
reported the opposite experience from these expectations, having a more positive train or bus
ride when they connected with a stranger than when they kept to themselves or had their typical
commute. These results suggest that even in places where connecting with strangers seems to be
actively discouraged, such as in Chicago or even London, connecting with strangers may be
more pleasant than the solitude people may be choosing to experience instead. However,
experiments testing exactly which aspects of social interaction people might misunderstand, from
starting a conversation to carrying it out to wrapping it up, are limited. People might
misunderstand the outcomes of social interactions because they misunderstand how interested
Hello Stranger, 5
others are in engaging with them in the first place, or because they misunderstand how an
interaction will unfold once it has started.
Here we report a larger-scale conceptual replication and extension of Epley and
Schroeder (2014) using a new (and perhaps more reserved) sample of British participants
commuting in and out of London. We made three meaningful changes to the original
experiments: we significantly increased our sample size, we measured expectations and
experiences within-participants rather than between-participants to enable direct comparisons
between them, and we asked participants in the conversation condition to simply talk with
another commuter rather than to “try to make a connection” and “get to know” a fellow
commuter.
This replication effort is valuable for testing the robustness of the original experimental
results in a different culture. Recent research using an experience sampling method found that
people who reported being in the presence of strangers also reported being less happy than they
were at the preceding measurement time, leading the authors to conclude “that interactions with
strangers are, on average, unpleasant” (Quoidbach et al., 2019, pg. 9). However, this experience
sampling research could not reliably assess whether people were actually interacting with these
strangers in a conversation or merely in the presence of strangers (which could be confounded
with other experiences that might be relatively unpleasant, such as sitting in a waiting room with
strangers at a doctor’s office). By experimentally manipulating whether people actually interact
with strangers in conversation or not, we can draw causal inferences about the consequences of
engaging with strangers for wellbeing. Experiment 1 also serves to test a potential alternative
interpretation of the original experiments. Specifically, by changing the experimental instructions
to simply ask participants to have a conversation, we test a possible alternative interpretation in
Hello Stranger, 6
the original experiments that connecting with strangers in public is pleasant only when explicitly
instructed to try to make a connection (see Quoidbach et al., 2019).
In Experiment 2, we report a novel test between two hypotheses about the psychological
barriers that might make people reluctant to connect with strangers even when it would be more
pleasant than what they might otherwise be doing (see also Sandstrom & Boothby, 2020). One
hypothesis is that people think that initially engaging a stranger will be relatively unpleasant
because social norms or observable cues suggest others are uninterested in talking. In contexts
without obvious cues that others are interested in engaging in conversations, such as on
commuter trains or other public spaces, uncertainty about others’ interest or willingness to
engage in conversation could keep people from trying to start a conversation out of fears of being
rejected or being impolite. This reluctance to try to start a conversation could then lead most
people to avoid even trying to engage a stranger in conversation, thereby creating the perception
that others are unwilling or uninterested in talking based on a tendency to infer that others’
mental states match their observed actions (i.e., the correspondence bias; Gilbert & Malone,
1995). Indeed, commuters on trains and busses in Chicago believed that they were personally
more interested in talking with other commuters than other commuters were in talking with them
(Epley & Schroeder, 2014; Experiment 3). Underestimating others’ interest in engaging in
conversation could create a barrier to starting conversations with strangers more often in daily
life.
A second hypothesis is that people think enacting a conversation with a stranger will be
potentially unpleasant, such that people think the conversation itself will be unpleasant and hence
worth avoiding. Only one experiment we know of has tested this hypothesis directly (Epley &
Schroeder, 2014; Experiment 3), and finds no significant evidence to support it. Participants in
Hello Stranger, 7
this experiment imagined having either a positive conversation with an interesting person, a
negative conversation with an uninteresting person, or simply a conversation with an unspecified
stranger. If participants expected that the conversation itself would be unpleasant, then those who
imagined having a conversation with an unspecified stranger would anticipate experiences closer
to the negative interaction condition than to the positive interaction condition. Instead, the results
from this experiment indicated that participants’ anticipated conversations with an unspecified
stranger were more similar to the positive interaction condition, suggesting that people might not
expect that actually having a conversation would be unpleasant.
To test between these two hypotheses, we ask people to predict how they would feel
when trying to have a conversation with a stranger compared to how they would feel actually
having a conversation. If the psychological barrier to talking with strangers stems from fears of
engaging others in conversation, then people should expect a more negative experience when
they imagine trying to have a conversation (because one could be rejected) than when they
expect to actually have a conversation. If, however, the barrier to talking with strangers stems
from anticipating a negative conversation, then people could expect similarly negative
experiences whether trying to talk or actually talking, or even anticipate a less negative
experience when trying to talk because a potentially unpleasant conversation might be avoided.
We report how we determined our sample size, all data exclusions, all manipulations, and
all measures in each study. Data files, surveys, and preregistrations for both experiments can be
found online [https://osf.io/cd86s/?view_only=8b3ad7213eb6457399f6cf066a82d24a].
Experiment 1: Talking in London
We asked train commuters either going into London (in the morning) or out from London
(in the evening) to talk with a fellow passenger on the train, sit in solitude, or do whatever they
Hello Stranger, 8
normally do. We hypothesized that commuters who talked with a fellow passenger would report
a more positive commute than those who sat in solitude or did what they normally do (which
casual observation confirms typically involves some amount of sitting in solitude). We also
hypothesized that participants’ expectations would be systematically miscalibrated, such that
people would expect that talking would be relatively less pleasant than keeping in solitude,
leading them to underestimate how positive it could be to connect with a stranger. In addition to
measuring mood, we measured how much participants reported learning on their commute and
how productive they found their commute to be, to assess additional benefits and costs that might
come from connecting with strangers. Finally, we measured contextual and dispositional factors
that could moderate participants’ expectations or experiences, including trait extraversion,
demographic characteristics, commute length, and normal commuting behavior.
Method
We preregistered our methods, predictions, and analysis plan for this experiment at
https://aspredicted.org/blind.php?x=ki6i24. Full procedural details are described in the
Supplemental Materials.
Participants. We targeted at least 300 participants, but continued collecting data through
scheduled shifts after we reached that point. In total, we recruited 466 individuals to participate
in exchange for a 10£ Starbucks gift card from four public transport stations in the London area:
Liverpool Street Station, Cambridge, Chelmsford, and Colchester. Of these, 383 completed the
full survey3 (82.19% response rate overall; Mage = 39.14 years, SD = 11.724; 60.31% male). The
3 One of these participants only completed the post-survey (not the pre-survey) but we include their post-survey in
the analysis. 4 Two participants did not report age; two participants in the analyzed sample reported being below 18 years old (15
years old and 17 years old).
response rate did not vary significantly across experimental conditions, χ2 (n = 466) = 3.07, p =
.215. See Supplemental Table S1 for sample characteristics by experimental condition.
Procedure. This experiment utilized a 2 (survey timing: pre-commute or post-commute)
× 3 (instruction condition: conversation, solitude, or control) experimental design with repeated
measures on the first factor.
Two or more experimenters stood at a train platform at one of four pre-determined
London public transportation stations between the hours 5:30 to 9:30 am (morning) and 4:00 to
7:00 pm (evening) recruiting participants for an experiment on commuting. The experimenters
recruited participants when the next train was at least 7 minutes out from arrival to ensure that
sessions would not be interrupted by incoming trains. The experimenter enrolled anyone who
agreed to participate after learning that it would require 5 minutes to complete an online survey
before and after their commute.
Once enrolled, experimenters randomly assigned participants into one of three
experimental conditions: conversation, solitude, or control. Experimenters asked participants to
take a seat away from the larger crowd of people waiting to ensure instructions could be
communicated effectively. Commuters received a printed card with their condition instructions
(see Supplemental Materials for full text), which the experimenter also read aloud to them to
ensure comprehension.
The conversation condition instructions were:
“On your commute this morning, please try to strike up a conversation with a fellow
commuter. You can talk to anyone you want about anything you wish. For example, you
might start by asking someone where they live and how long they have lived there, what
they do for a living, or what they think about a particular news story. You can then give
background about yourself, where you live, what you do for a living, or what you think
about a particular news story. Try in whatever time you have to get to know this person a
little bit, and let them get to know you a little. You can talk to this person for as long or
as little as you would like, although it would be best if you could continue the
Hello Stranger, 10
conversation for as long as the conversation naturally allows. You should try to use your
commute this morning to have a conversation.”
The solitude condition instructions were:
“On your commute this morning, please do not engage anyone in conversation. Instead,
take this time to sit alone with your thoughts. You can think about whatever you have
going on in the day, or just let your mind wander. You should try to use your commute
this morning to have some time alone.”
The control condition instructions were:
“On your commute this morning, please do whatever you would normally do. You may
choose to talk, write, read, work, sit quietly, sleep or any other activity you typically do.
You should try to use your commute this morning to do what you would normally do.”
The experimenter then asked if the participant was comfortable following the
instructions. If a participant indicated they were not comfortable, the experimenter reiterated the
importance of following the instructions but allowed them to leave the experiment if they no
longer wanted to continue. Overall, 5.58% of commuters dropped out at this point (11.18% in the
conversation condition, 4.14% in the control condition, 1.25% in the solitude condition).
Participants then opened the initial survey on their phone before starting their commute.5
When finished, the experimenter asked participants to keep the survey open so they could
complete the second half after their commute. Participants were told they would receive half of
their compensation for completing the first survey and the rest for completing the second survey.
The experimenter asked for the participant’s phone number in order to send a text reminder to
complete the second survey. Participants then boarded their train when it arrived and completed
the experiment on their own.
Pre-commute survey. Participants first reported their normal commuting behavior:
“What do you normally do during your commute? Please mark all options that apply” (talk to
someone, talk/socialize on a computer/phone, read, sleep, think (by yourself), work, or other).
5 Some participants (2.35%) did not have a phone. These participants used the experimenter’s phone instead and
received a link for where they could complete the survey when they were at a computer.
Hello Stranger, 11
Participants then reported their expectations about their commute on five items: 1. How
happy do you think you will feel after your commute today, compared to normal?; 2. How sad do
you think you will feel after your commute today, compared to normal?; 3. How pleasant will
your commute be today, compared to normal? 4. How much do you think you will learn on your
commute today, compared to normal? 5. How productive will you be during your commute
today (that is, how much will you accomplish), compared to normal? The response scale for the
happy, sad, pleasant, and productive items included 7 scale points that ranged from -3
(Significantly less happy/sad/pleasant/productive) to 3 (Significantly more
happy/sad/pleasant/productive). The response scale for the learning item included 7 scale points
that ranged from -3 (Learn significantly less) to 3 (Learn significantly more).
Commuters then completed the 8-item extraversion subscale from the Big 5 Personality
Inventory (John, Donahue, & Kentle, 1991; e.g., “I see myself as someone who is talkative”; “I
see myself as someone who is full of energy,” α = .876) on 5-point scales ranging from -2
(strongly disagree) to 2 (strongly agree), and entered demographic information (gender, age, and
ethnicity).
Post-commute survey. Participants first reported their actual commuting behavior:
“What did you actually do during your commute? Please mark all options that apply” (talked to
someone, talked/socialized on a computer/phone, read, slept, thought (by yourself), worked, or
other). Participants then reported their actual experience on the same five items measured in the
pre-commute survey on the same response scales.
Participants reported demographic information about one of their fellow commuters: 1.
Age (older than me, younger than me, about the same age as me, not sure); 2. Gender (female,
male, not sure); 3. Race (same race as me, different race than me (please report), not sure).
Hello Stranger, 12
If participants reported talking to someone, they reported these demographic features for “the
person with whom you spoke today.” If participants reported that they did not talk to someone,
they reported these demographic features about “the person who sat closest to you today.”
We also asked the participants who reported talking to someone four items about their
conversation experience: 1. For approximately how long did you speak to the person? (free
response); 2. What did you talk about? (free response); 3. How interesting was the conversation
with the person? (0 = not at all interesting; 4 = very interesting); 4. What is your overall
impression of the person? (-3 = very negative; 3 = very positive).
Finally, the post-commute survey included several exploratory questions. Participants
reported how likely they were “to have a conversation with someone on [their] commute
tomorrow” (0 = not at all likely; 4 = very likely), whether someone sat next to them during their
commute (1 = yes, 2 = no; sample M = 1.30, SD = 0.46), how the commute made them feel (free
response), and the duration of their commute (free response).
Results
instructions, although this figure varied by experimental condition such that fewer reported
following instructions in the conversation condition (78.0%) than in the solitude (97.1%) and
control (100%) conditions, χ2 (n=383) = 47.22, p < .001).6 All of the following analyses include
all participants in each condition, regardless of whether they reported following the instructions
or not (i.e., intent-to-treat analyses). Reported effects are consistently larger when analyses are
6 We computed how many participants followed the commuting instructions that they were given based on what
they reported actually doing during their commute (i.e., the first item in the post-commute survey). Participants who
reported “talking to someone” were counted as following the instructions in conversation condition but as not
following the instructions in the solitude condition; we coded all participants in the control condition to have
followed instructions.
See Supplemental Materials).
Positivity. As preregistered, we computed the expected and experienced commute
positivity by averaging happiness, sadness (reverse-scored), and pleasantness into a composite
variable (α = .85). A 2 (survey timing: pre-commute vs. post-commute) × 3 (condition:
conversation, solitude, or control) mixed model ANOVA on positivity yielded significant main
effects of survey timing, F(1, 379) = 5.19, p = .023, ηp 2 = 0.01, and condition, F(2, 379) = 16.63,
p < .001, ηp 2 = 0.08, qualified by the predicted interaction, F(2, 379) = 3.09, p = .047, ηp
2 = 0.02
(Figure 1).
Consistent with our hypotheses, participants’ actual experiences varied by experimental
condition, F(2, 379) = 12.59, p < .001, ηp 2 = 0.06. Specifically, participants reported having a
significantly more positive commute in the conversation condition (M = 0.89, SD = 0.98) than in
either the solitude condition (M = 0.35, SD = 0.97), t(263) = 4.44, p < .001, d = 0.55, or the
control condition (M = 0.41, SD = 0.91), t(243) = 3.92, p < .001, d = 0.50. The difference
between the solitude and control conditions was nonsignificant, t(254) = -0.50. These findings
appear to be robust: In an analysis that controlled for participant extraversion, participant and
partner demographic matches (same age, gender, and race), commute duration in minutes, and
normal commuting behaviors, the effect of being in the conversation condition on rated positivity
of actual commute remained statistically significant compared to the solitude and control
conditions, ps < .001.
Inconsistent with our hypotheses, participants also expected the same overall pattern of
results across conditions before their commute, F(2, 379) = 13.62, p < .001, ηp 2 = 0.07, expecting
to have a more positive commute in the conversation condition (M = 0.73, SD = 0.90) than in the
Hello Stranger, 14
solitude condition (M = 0.40, SD = 0.79), t(263) = 3.11, p = .002, d = 0.38, or the control
condition (M = 0.19, SD = 0.71), t(243) = 5.06, p < .001, d = 0.65.
The significant interaction between survey timing and condition on positivity indicates
that participants’ expectations differed from their actual experiences. As predicted, participants
significantly underestimated how positively they would feel in the conversation condition, paired
t(126) = -2.04, p = .047, d = -0.17. Unexpectedly, participants also significantly underestimated
how positive they would feel in the control condition, paired t(117) = -2.38, p = .019, d = -0.27.
Expectations did not differ significantly from reported experiences in the solitude condition,
paired t(137) = 0.69, p = .490, d = 0.06. Although we observed the predicted interaction between
survey timing and condition on positivity, with participants in the conversation condition
underestimating the positivity of their commute, the interaction did not reflect the hypothesized
pattern of results because participants did not expect to have a less positive experience in the
conversation condition than in the solitude and control conditions.
Productivity and learning. We conducted the same 2 × 3 ANOVAs on predicted and
actual productivity and learning during the commute. Consistent with results reported in Epley
and Schroeder (2014), we observed no significant main effects or interaction on productivity,
Fs(2, 379) < 2, ps > .15, ηp 2 = 0.12, such that participants both expected and reported learning
more in the conversation condition than in the control condition, t(243) = 4.20, p < .001, d =
0.54, or the solitude condition, t(263) = 5.88, p < .001, d = 0.73. Neither expected nor reported
learning differed between the control and solitude conditions, t(254) = -1.47, p = .143, d = 0.18.
Hello Stranger, 15
Figure 1. Predicted and actual positivity of commute (top panel), learning during commute
(middle panel), and productivity of commute (bottom panel) by experimental condition (control,
solitude, conversation) in Experiment 1. Error bars represent standard error around the mean.
0
0.2
0.4
0.6
0.8
1
1.2
0
0.2
0.4
0.6
0.8
1
1.2
-0.2
0
0.2
0.4
0.6
Hello Stranger, 16
Likelihood of future conversations. Only 14.4% of participants in this experiment,
across conditions, reported regularly talking to other commuters on the train. If misunderstanding
the consequences of social interaction creates a barrier to talking with others more often, then
those who have conversations should report being more interested in having conversations in the
future compared to those who do not have conversations. Consistent with this possibility, we
observed a significant effect of condition on the reported likelihood of having a conversation on
their commute the next day, F(2, 380) = 17.76, p < .001, ηp 2 = 0.09, such that participants in the
conversation condition reported being more likely to talk again (M = 1.46, SD = 1.23) than those
in the solitude condition (M = 0.89, SD = 1.27), t(263) = 3.73, p < .001, d = 0.46, and the control
condition (M = 0.58, SD = 1.0), t(243) = 6.11, p < .001, d = 0.78. Unexpectedly, participants in
the solitude condition also reported being more likely to talk the next day than those in the
control condition, t(254) = 2.13, p = .034, d = 0.27, perhaps because we asked them to explicitly
avoid having a conversation on the day of the study itself and hence may have expected to be
slightly more likely to talk the following day.
Exploratory analyses. We conducted five exploratory analyses on the commuters who
reported actually talking (N = 99) to better understand the nature of the conversations.
Conversation experiences and personal characteristics. First, to better understand
participants’ experiences talking with a stranger, we examined the correlations between actual
commute experiences (positivity, learning, and productivity), predicted commute experiences on
the same variables, and the misprediction of experiences with characteristics of participants
(extraversion), characteristics of participants’ conversations (e.g., length, interestingness), and
participants’ demographics (gender, age, race) among the commuters who reported talking. As
shown in Table 1, expected and experienced positive mood in the conversation condition was
Hello Stranger, 17
more positive impressions of the conversation partner. The misprediction of positive mood was
only associated with longer conversations, more interesting conversations, and more positive
impressions of the conversation partner. Actual learning was positively associated with longer
and more interesting conversations, more positive impressions of the conversation partner, and a
larger age gap between conversationalists, whereas expected learning was associated with more
interesting conversations and higher impressions of partner, and mispredictions in learning were
associated with longer and more interesting conversations. Finally, expected and experienced
productivity was not associated with any of the measured variables. In general, the longer
participants’ conversations, the more they underestimated their positive experience.
Table 1
Partner Characteristics Among Participants Who Reported Talking During their Commute in
Experiment 1 (N = 99)
Positivity (Diff Score) .014 -.266** -.238* -.263** .022 .099 .196
Predicted – Experienced
Learning (Diff Score) -.033 -.356** -.319** -.103 .139 .011 .144
Predicted – Experienced
Productivity (Diff Score) .129 .086 .027 -.002 -.179 -.083 -.048
Predicted Positivity
Predicted Learning (Pre-
Predicted Productivity
Experienced Positivity
Hello Stranger, 18
Experienced
Productivity
(Post-Commute)
Notes. *<.05, **<.01, ***<.001. A higher difference score means more overestimation (or less
underestimation) of positivity, learning, or productivity. We additionally examined correlations
between experienced and predicted positivity, learning, and productivity pre- and post-commute
among extraversion levels of participants in the control and solitude conditions; no correlations
achieved statistical significance at the p < .05 level.
Conversation partners. Second, we tested which variables predicted partner
characteristics (i.e., to whom participants talked) in a series of linear regression analyses. There
were no effects of participants’ own demographics or extraversion on the demographics of their
conversation partner except that female participants were more likely to talk with other females
(73%) than males, whereas male participants were similarly likely to talk with females (50.8%)
and males, η2(1, 98) = 4.68, p = .031.
Gender. Third, to identify any effects of gender, we conducted a series of 2 (own gender:
male, female) × 2 (partner’s gender: male, female) ANOVAs on predicted and experienced
positivity, learning, and productivity of commute after talking. No statistically significant effects
emerged (ps > .110) except that females predicted that they would learn more (M = 1.26, SE =
0.16) than males (M = 0.82, SE = 0.11), F(1, 94) = 4.99, p = .028, ηp 2 = 0.05.
Selection. Fourth, to examine potential selection effects between those who reported
following our instructions (N = 99) and those who did not (N = 28) in the conversation condition,
we compared these two groups on predicted positivity, learning, and productivity, participant and
partner demographic characteristics, participant extraversion, and commute duration. No
statistically significant effects emerged on any of these variables, ts(125) < 1.92, ps > .057, ds <
0.34, except for target gender, whereby participants who talked were more likely to talk with a
female (M = 0.59, SD = 0.50) whereas the people who did not talk were more likely to report
Hello Stranger, 19
sitting next to a male (M = 0.29, SD = 0.46), t(125) = 2.89, p = .005, d = 0.52. These results
suggest that following instructions in the conversation condition was not driven by
characteristics about the participants themselves but rather may have been driven the context
these participants found themselves in, such as sitting next to a man or potentially not sitting
close to anyone at all. In their open-ended comments, those who did not follow instructions in
the conversation condition reported difficulty finding someone to talk to, rather than being
rejected when they tried to talk with someone. Difficulty finding a conversation partner could
reflect an uncrowded train, or mistaken perceptions of others’ interest in talking if the participant
had tried (such as from seeing others wearing headphones; see Appendix A for participants’ own
explanations). We could not conduct similar analyses in the solitude condition because few
participants reported not following instructions.
Finally, additional analyses indicated participants who followed instructions in the
conversation condition reported experiencing a more positive commute (M = 1.02, SD = 1.02)
than participants who did not follow the instructions (M = 0.37, SD = 0.62), t(125) = 3.21, p =
.002, d = 0.57, and also reported learning more after their commute (M = 0.96, SD = 0.97) than
participants who did not follow the instructions (M = 0.36, SD = 0.56), t(125) = 3.14, p = .002, d
= 0.56. We observed a nonsignificant difference in reported productivity between those who
followed instructions in the conversation condition and those who did not, p = .598. Even within
the conversation condition, those who talked with a stranger had a more positive commute than
those who, for whatever reason, did not. Note that those who did not follow instructions in the
conversation condition experienced commutes more similar, on average, to those in the solitude
and control conditions who likewise were typically not talking with others.
Discussion
Hello Stranger, 20
Consistent with prior research, London-area train commuters, who do not have a
reputation for being especially social, nevertheless reported having a more positive commute
after being randomly instructed to strike up a conversation with a stranger than after being asked
to keep to themselves in solitude or do whatever they normally do. We observed no
corresponding negative consequences of experiencing a less productive commute. Inconsistent
with prior research, however, commuters who were instructed to imagine talking to a stranger
did not anticipate a less positive commute than those in the control or solitude conditions.
Although participants in the conversation condition significantly underestimated how positive
their commute would be, they also expected that it would be more positive overall than their
commute in the solitude and control conditions. This is surprising both because this result
diverges from prior research, and because conversations between strangers are so rare on these
commuter trains.
One potentially important difference in the instructions from Epley and Schroeder (2014)
is that participants in the current experiment imagined successfully following the instructions
that they had been given and then reported how they expected to feel following the experience.7
For instance, in the conversation condition, participants imagined having a conversation with a
fellow commuter and then reported how they expected to feel afterwards. In contrast, the
instructions in Epley and Schroeder (2014) asked commuters to imagine trying to enact the
experimental instructions (e.g., “imagine trying to have a conversation”) and then to report how
they expected to feel, in each of the three experimental conditions. We theorized that trying to
have a conversation calls to mind the act of engaging someone to start a conversation, whereas
successfully following the instructions would focus more on the enactment of the conversation
7 This change to the instructions was unintentional. Without noticing, we worked from a file containing an early
version of instructions rather than the final instructions actually used in the experiments.
Hello Stranger, 21
itself. Prior research (Epley & Schroeder, 2014, Experiment 3) indicated that people might be
reluctant to connect with strangers because they underestimate how interested others are in
talking with them, rather than because they imagine having an unpleasant conversation with
another person. If initially engaging others in conversation is the key psychological barrier that
creates reluctance to talk with strangers, then instructing participants to imagine trying to talk
could lead them to expect a more negative experience than instructing them to imagine actually
talking. We test this possibility in Experiment 2.
Experiment 2: Predictions About Trying to Converse or Successfully Conversing
Commuters in Experiment 2 reported how they expected to feel after either trying to
follow the experimental instructions from Experiment 1, or successfully following the
instructions. We predicted that participants in the solitude and control conditions would
anticipate no difficulty enacting these instructions because they require no coordination with
others. In contrast, we hypothesized commuters in the conversation condition would expect a
worse experience when they imagined trying to have a conversation than when they imagined
actually having a conversation. This result would suggest that concerns about starting a
conversation may be a more significant barrier to engaging in conversation with a stranger than
concerns about how pleasant or enjoyable the conversation will be.
Method
We preregistered our methods, predictions, and analysis plan for this experiment at
https://aspredicted.org/blind.php?x=ri2gx9.
Participants. We collected surveys from the same four train stations used in Experiment
1. We targeted a total sample of 200 participants over the four days we were allowed to conduct
Hello Stranger, 22
this experiment, but were unable to reach this goal. In total, 148 people completed the survey
(Mage = 38.11 years, SD = 11.67; 56.76% male) in exchange for 5£.
Design. This experiment utilized a 2 (instructions: try or succeed) × 3 (condition:
conversation, solitude, or control) experimental design with repeated measures on the second
factor.
Procedure. We recruited participants using the same method as Experiment 1, except
that we told participants that the study only required completing one survey during their
commute. We provided participants with the survey link, which they opened on a digital device
with the experimenter present. Participants completed the consent form and the extraversion
scale from Experiment 1 (α = .867) on the train platform, and then signed a form promising to
complete the rest of the survey during their commute in exchange for a 5£ gift card (which we
sent to them after completing the survey).
Survey. Participants first indicated what they “normally do during [their] commute” by
marking any of the following options that applied: talk to a stranger, talk to an acquaintance or
friend, talk/socialize on a computer/phone, read, sleep, think (by yourself), work, or other
(presented in a randomized order).8 Participants then reported their expectations about “what it
would be like to talk to a fellow commuter” on 5 different items: 1) How interested would you be
to talk to another fellow commuter? (1 = not at all interested, 7 = very interested), 2) How
interested would another fellow commuter be to talk to you? (1 = not at all interested, 7 = very
interested), 3) How difficult would it be to start the conversation? (1 = not at all difficult, 7 =
very difficult), 4) How difficult would it be to end the conversation? (1 = not at all difficult, 7 =
8 If participants reported that they normally talk to a stranger or acquaintance/friend we asked them to “please
estimate how often you do it” (free response). If participants checked “other”, we asked them to “please provide
details about what you do.”
Hello Stranger, 23
very difficult), and 5) What percentage of people riding the train do you think would be willing to
talk to you? (enter percentage).
Participants then read a short description of the procedure used to recruit participants on
the train platform in Experiment 1 to help them simulate the experience, and then read one
randomly selected set of instructions from either the conversation, solitude, or control conditions
from Experiment 1. Participants randomly assigned to the succeed condition imagined that they
“successfully follow these instructions,” whereas participants in the try condition imagined that
they “try to follow these instructions.” Participants then predicted how positive and productive
their commute would be, and how much they would learn, using adapted items from Experiment
1: 1) How happy would you feel after your commute today, compared to normal, [after
successfully following] / [if you tried to follow] these instructions? (-3 = significantly less happy,
3 = significantly more happy), 2) How sad would you feel after your commute today, compared
to normal, [after successfully following] / [if you tried to follow] these instructions? (-3 =
significantly less sad, 3 = significantly more sad), 3) How pleasant would your commute be
today, compared to normal, [after successfully following] / [if you tried to follow] these
instructions? (-3 = significantly less pleasant, 3 = significantly more pleasant), 4) How much
would you learn on your commute today, compared to normal, [after successfully following] / [if
you tried to follow] these instructions? (-3 = learn significantly less, 3 = learn significantly
more), 5) How productive would you be (that is, how much would you accomplish) during your
commute today, compared to normal, [after successfully following] / [if you tried to follow]
these instructions? (-3 = significantly less productive, 3 = significantly more productive).
Participants then viewed each of the remaining two sets of instructions (from the conversation,
solitude, or control conditions) and reported their expectations in each condition.
Hello Stranger, 24
Results
Positivity. As in Experiment 1, we averaged predicted happiness, sadness (reverse-
scored), and pleasantness to make an index of commute positivity (α = .80). A 2 (instructions: try
or succeed) × 3 (condition: conversation, solitude, or control) mixed model ANOVA on
predicted commute positivity yielded a significant main effect of condition, F(2, 292) = 14.49, p
< .001, ηp 2 = 0.09, and a significant main effect of instructions, F(1, 146) = 4.03, p = .047, ηp
2 =
0.03, qualified by the predicted interaction, F(2, 292) = 3.64, p = .027, ηp 2 = 0.02 (see Figure 2).
This interaction indicates that the experimental instructions only influenced evaluations in the
conversation condition, such that participants expected a significantly more positive commute
when imagining that they successfully followed the instructions (M = 1.00, SD = 1.22) than
when they imagined trying to follow the instructions (M = 0.48, SD = 1.24), t(146) = 2.58, p =
.011, d = 0.43. There was no effect of instructions in the solitude or control conditions, ts(146) =
-.07 and .51, respectively.
Additional analyses indicated that participants in the succeed condition expected a more
positive experience in the conversation condition than in the solitude condition (M = 0.21, SD =
0.85), paired t(71) = 4.42, p < .001, d = 0.75, and in the control condition (M = 0.28, SD = 0.60),
paired t(71) = 4.53, p < .001, d = 0.75, with a nonsignificant difference in expectations between
the solitude and control conditions, paired t(71) = -0.63. In contrast, participants in try condition
did not expect a more positive experience in the conversation condition than in the solitude
condition (M = 0.22, SD = 1.01), paired t(75) = 1.36, p = .179, d = 0.23, or the control condition
(M = 0.24, SD = 0.50), paired t(75) = 1.63, p = .017, d = 0.25, with a nonsignificant difference in
expectations between the solitude and control conditions, paired t(75) = -0.12.
Hello Stranger, 25
Productivity and learning. We conducted the same 2 × 3 ANOVA on expected learning
and productivity (see Figure 2). For expected learning, we observed only a significant main
effect of condition, F(2, 292) = 72.74, p < .001, ηp 2 = 0.33, such that participants expected to
learn more in the conversation condition (M = 1.20, SD = 1.21) than in the solitude condition (M
= -0.09, SD = 1.17) or control condition (M = 0.27, SD = 0.72). For expected productivity, we
also observed a significant main effect of condition, F(2, 292) = 7.04, p = .001, ηp 2 = 0.05, such
that participants expected their commute to be least productive in the conversation condition (M
= -0.13, SD = 1.47) compared to the control condition (M = 0.34, SD = 0.74) and solitude
condition (M = 0.06, SD = 1.26). We additionally observed a marginally significant interaction
between condition and instructions on expected productivity, F(2, 292) = 2.79, p = .063, ηp 2 =
0.02, such that the effect of the instructions was somewhat larger in the conversation condition
than in the solitude or conversation condition.
Hello Stranger, 26
Figure 2. Predicted positivity of commute (top panel), learning on commute (middle panel), and
productivity of commute (bottom panel) by experimental condition in Experiment 2. Error bars
represent standard error around the mean.
-1
0
1
2
Hello Stranger, 27
Consistent with the results of Epley and Schroeder (2014) suggesting that people may be
reluctant to talk with others because of concerns about starting a conversation rather than about
the outcome of the conversation itself, we found that participants believed they were more
interested in talking (M = 2.57, SD = 1.92) than others were (M = 2.28, SD = 1.83), paired t(147)
= 2.41, p = .017, d = 0.15, that participants believed it would be more difficult to start a
conversation (M = 3.43, SD = 1.76) than to end a conversation (M = 2.53, SD = 1.89), paired
t(147) = 5.15, p < .001, d = 0.49, and that participants believed that relatively few of their fellow
commuters (M = 25.39%, SD = 22.19%) would be willing to talk with them (significantly less
than 50%, as we had hypothesized, one-sample t(147) = -13.49, p < .001).
Exploratory analyses. We conducted two additional analyses that we preregistered as
exploratory.
examined associations between expectations about starting a conversation and the expected
positivity, productivity, and learning on one’s commute in the try and succeed conditions (see
Table 2). Correlational analyses indicate that participants’ own interest in talking and their
perceptions of others’ interest in talking are positively related to the expected positivity, amount
of learning, and productivity of the commute (rs > .138). Contrary to our predictions, we did not
observe consistent relationships between the perceived difficulty of starting or ending a
conversation, the expected percentage of others willing to talk, and expected experiences talking
with strangers.
Correlations Between Interest in Talking, Difficulty Starting and Ending the Conversation, and
Expected Commute Experiences (Positivity, Learning, Productivity) in the Conversation
Condition When Participants Imagine Trying to vs. Succeeding in Talking (Experiment 2)
Own
Interest
in
Talking
Perceived
Learning on Commute .326** .138 .093 .149 .002
Productive Commute .274* .164 .009 .118 .144
Succeed Condition
Learning on Commute .243* .304** -.065 -.092 -.104
Productive Commute .383** .269* -.052 -.252* .046
Note. * p < .05, ** p < .01
Hello Stranger, 29
examined associations between participant characteristics (age, gender, extraversion, and normal
behavior on trains) and expected commute experiences in each experimental condition. As
shown in Table 3, few significant correlations emerged with age and gender. Significant
correlations, however, did emerge with people’s past experiences talking with strangers and their
natural inclination to be more extroverted. If people are overly reluctant to connect with
strangers because they underestimate the extent to which others are interested in talking to them
(Epley & Schroeder, 2014), then those who have more actual experience talking with strangers
should also have more calibrated expectations, meaning that they should also expect more
positive experiences when they imagine trying to connect with a stranger. Consistent with this
possibility, the more people report normally talking to strangers, the more they also expected a
positive experience when trying to implement the conversation instructions (r = .248, p = .031).
However, people expected a more negative commute trying to implement the solitude instruction
(r = -.376, p = .001).
And Expected Commute Experiences (Positivity, Learning, Productivity) by Condition in
Experiment 2
Learning on Commute .397** -.169 .262* .060
Productive Commute -.003 -.187 -.190 -.005
Solitude Condition
Learning on Commute .111 -.190 -.135 -.176
Productive Commute .030 -.045 -.059 .008
Control Condition
Learning on Commute -.048 .054 -.016 .169
Productive Commute -.095 .035 .149 .069
Succeed Condition
Conversation Condition
Learning on Commute .155 -.014 .081 -.045
Productive Commute .137 -.016 .030 .027
Solitude Condition
Learning on Commute .027 -.179 -.005 -.322**
Productive Commute .034 -.130 .021 -.313**
Control Condition
Learning on Commute .072 -.229 .090 -.168
Productive Commute .109 -.329** .073 -.122
Note. + p < .10, * p < .05, ** p < .01. Gender is coded 1 = male, 2 = female.
Hello Stranger, 31
Discussion
People imagined that trying to talk with a stranger to be a less positive experience than
imagining actually talking with a stranger, suggesting that a barrier to social engagement may
come not from expecting a poor interaction with others but rather from anticipating that others
are relatively disinterested in engaging in conversation. This seems to occur, at least in part,
because people expect that strangers are relatively disinterested in talking with them. Consistent
with this possibility, people who were more experienced with starting conversations (e.g.,
extraverts, people who normally talk with strangers) did not anticipate as big a difference in
outcomes when trying versus successfully having a conversation. Experiment 2 further revealed
some reasons why people may expect starting conversation to be unpleasant: people thought they
would be more interested in talking than others would be, expected that starting conversations
would be more difficult than ending conversations, and expected that only a minority (about
25%) would be willing to talk with them.
General Discussion
Human beings are deeply social and yet at times can appear deeply unsocial, avoiding
connecting with strangers even when it would be relatively easy to do so. Our findings suggest
that social disconnection in these settings does not stem from people maximizing their own
wellbeing by keeping to themselves. On trains going in and out of London, commuters randomly
asked to have a conversation with a stranger had a more positive experience, and reported being
more likely to talk to a stranger again in the future, than those randomly assigned to sit in
solitude or do whatever they normally do on their commute. Instead, our research suggests that
people may avoid pleasant conversations with others in situations when they perceive that others
do not want to connect with them. In Experiment 2, commuters expected that trying to have a
Hello Stranger, 32
conversation with another person would be less pleasant than actually having a conversation
with another person. This result appears to stem from concerns about starting conversations in
this context. Participants believed that they would personally be more interested in talking than
other commuters were, consistent with a form of pluralistic ignorance (Epley & Schroeder, 2014;
Prentice & Miller, 1996), and estimated that only a small minority (25%) of fellow commuters
would be willing to talk.
We believe these fears likely reflect a tendency to significantly underestimate others’
sociality, creating a misplaced barrier to engaging others in conversation. Although we do not
know the exact percentage of riders on our participants’ trains would be willing to have a
conversation, 78% of participants in the conversation condition of Experiment 1 reported having
a conversation of some kind that ranged from 1-60 minutes long. Based on their written
comments, those who reported not having a conversation were not rejected by others but instead
seemed more likely to infer that others did not want to talk hence did not try (see Appendix A).
Notice these perceptions that lead to social avoidance are likely to be self-fulfilling: believing
that others are unwilling to connect could keep people from trying and thereby learning that their
concerns about talking with strangers could be miscalibrated. Indeed, the more people reported
talking to strangers in daily life in Experiment 2, the less we observed a difference in
expectations about trying versus actually having a conversation, presumably because those who
frequently try to talk with others know that trying to talk is likely to result in actually having a
conversation (see also Sandstrom & Boothby, 2020). This result is consistent with a previous
finding (Epley & Schroeder, 2014; Experiment 4) that travelers riding in cabs who report
normally talking to their cab drivers also expect that doing so will be pleasant, consistent with
Hello Stranger, 33
these riders’ actual experiences, while those who report rarely talking to their cab drivers
mistakenly expect that talking will be less pleasant than keeping to themselves.
These experiments add to the growing literature identifying increased wellbeing from
expanding one’s social circle to connect with others even in relatively fleeting exchanges (e.g.,
Gunaydin et al., 2020; Sandstrom & Dunn, 2014). Perhaps more important, they document a
robust positive effect of connecting with strangers even in a culture well-known for being
reserved, in experiments that manipulate whether people are actually interacting with another
person in conversation or not. These experiments provide clear causal evidence that talking with
strangers in conversation is relatively pleasant (cf., Quoidbach et al., 2019). These experiments
also provide additional evidence about the precise barriers that might keep people from
connecting with strangers more often in daily life. London train riders did not seem to think that
talking with others would be systematically unpleasant as long as they could have a
conversation. Instead, they seemed to fear that others would be unwilling to talk in the first
place.
This mechanism predicts that people’s expectations of their enjoyment of a conversation
will be more calibrated when they are aware that another person with whom they would like to
talk is also interested in talking with them, compared to cases when there is more uncertainty
about another person’s interest in talking. Consistent with this possibility, people’s expectations
about their experience in conversations with friends (whose interest in talking can be presumed)
tend to be more calibrated than their expectations in conversations with strangers (Dunn et al.,
2008; Kardas, Kumar, & Epley, 2021). In another experiment, students in a dining hall were
more likely to talk with a stranger when they were able to wear wristbands indicating whether or
not they were interested in having a conversation, thereby signaling their interest clearly and
Hello Stranger, 34
enabling people to talk with others they knew were also interested in talking with them (see Lok
& Dunn, 2021). People should also be better calibrated predicting their experiences when they
know they are going to have a conversation with someone else, and hence are unlikely to be
rejected, compared to when there is more uncertainty about whether one’s attempt to start a
conversation will be successful. More direct tests of this hypothesis are needed, but one series of
experiment found that people about to have a conversation with a stranger in the lab did not
underestimate how much they would enjoy the opening moments of the conversation (but did
underestimate how much they would continue to enjoy repeated periods of conversation with the
same person; Kardas, Schroeder, & O’Brien, 2021).
Given these results, we predict the psychological barriers to engaging in conversations
with strangers are likely to vary across contexts based on the perceived risks of social interaction.
When another’s interest in connecting is uncertain, the perceived risks of trying to engage
another person are likely to dominate decision making (Murray, Holmes, & Collins, 2006). In
much of modern life, one cannot escape the constant presence of smartphones when out in public
or when traveling. Existing research has documented how smartphones may impose social costs
by distracting attention while in conversation, or by serving as a substitute for social interaction
altogether (Kushlev, Dwyer, & Dunn, 2019). We believe our research suggests another potential
social cost of smartphone technology: providing an imperfect signal that one is uninterested in
talking to others. Although donning headphones or staring into a phone could be used
deliberately to avoid conversations, it is also a mindless activity that might at other times lead
people to underestimate others’ interest in talking. As one commuter from Experiment 1 wrote,
“I looked at everyone on their phones with their earplugs in and felt awkward thinking about
striking up a conversation. So I chickened out. [N]ow I feel frustrated and sad” (See Appendix A
Hello Stranger, 35
for full quote). Understanding how these cues could be guiding, or systematically misguiding,
social behavior is an important topic for future research.
Once a person knows they are going to connect with another person in a given situation,
additional perceived risks may guide social decision making. Here, too, emerging research
suggests that people may underestimate the positive impact of reaching out and connecting with
others by underestimating how competently they will be able to carry on conversation with a
stranger (Sandstrom & Boothby, 2020), how interested others will be in discussing deep and
meaningful content (Kardas, Kumar, & Epley, 2021), how much others will like them after the
conversation (Boothby, Cooney, & Clark, 2018), how much they will learn (Atir, Wald, &
Epley, 2021), or how positively others will feel (Kumar & Epley, 2018; Boothby & Bohns, 2020;
Zhao & Epley, 2021). As with many decisions under uncertainty, potential risks are easy to
overestimate, potentially creating barriers in this case to connecting with others more often in
daily life.
Finally, it is possible that either the perceived or actual consequences of social
interactions could vary across cultures, helping to explain differences in social norms across
cultures. Based on our own data, train riders in Chicago and London rarely talk to strangers, but
you might be hard pressed to avoid talking with a stranger in other parts of the world. Variance
in the observed sociality of people around the world could stem either from differences in the
actual experience of connecting with others, or from differences in people’s beliefs or
expectations about the consequences of connecting with others. The data we report here suggests
cultural differences in observed behavior may be more likely to stem from differences in
people’s expectations about the outcomes of conversations, which could then drive behavior and
create strong social norms. Underestimating how interested others are in having a conversation
Hello Stranger, 36
could make people reluctant to try to start one, thereby keeping people from obtaining the very
experience in their daily life that would serve to create more calibrated expectations. In one
experiment with cab riders in Chicago (Epley & Schroeder, 2014, Experiment 4), those who
reported routinely having conversations with their drivers expected that having a conversation
would be more pleasant than those who report rarely talking with their driver, even though both
groups of cab riders had a more pleasant commute when they talked to their driver than when
they did not. The newsworthy resistance some Londoners showed to the possibility of talking
more on their commuter trains stands in stark contrast to the actual experience of the Londoners
in our experiments who were significantly happier when encouraged to talk with another person,
and who then expressed more interest in talking with another person on a future trip. It is
possible that those living in more reserved cultures would enjoy talking with strangers every bit
as much as those living in more sociable cultures, but simply have fewer social interactions from
which they could learn just how much they would enjoy those conversations.
Hello Stranger, 37
References
Atir, S., Wald, K., & Epley, N. (2021). Conversation is surprisingly informative. Manuscript in
preparation.
Boothby, E. J., & Bohns, V. K. (2020). Why a simple act of kindness is not as simple as it
Seems: Underestimating the positive impact of our compliments on others. Personality
and Social Psychology Bulletin. https://doi.org/10.1177/0146167220949003
Boothby, E. J., Cooney, G., Sandstrom, G. M., & Clark, M. S. (2018). The liking gap in
conversations: Do people like us more than we think?. Psychological Science, 29(11),
1742-1756. https://doi.org/10.1177/0956797618783714
Boyce M.W. (2012) British Masculinities: Duty, confinement, and stiff upper lips. In: The
Lasting Influence of the War on Postwar British Film. Palgrave Macmillan, New York.
Dunn, E. W., Biesanz, J. C., Human, L. J., & Finn, S. (2007). Misunderstanding the affective
consequences of everyday social interactions: The hidden benefits of putting one’s best
face forward. Journal of Personality and Social Psychology, 92, 990 –1005.
Epley, N., & Schroeder, J. (2014). Mistakenly seeking solitude. Journal of Experimental
Psychology: General, 143, 1980–1999.
Fleeson, W., Malanos, A. B., & Achille, N. M. (2002). An intraindividual process approach to
the relationship between extraversion and positive affect: Is acting extraverted as “good”
as being extraverted? Journal of Personality and Social Psychology, 83, 1409 –1422.
Gunaydin, G., Oztekin, H., Karabulut, D. H., & Salman-Engin, S. (2020). Minimal social
interactions with strangers predict greater subjective well-being. Journal of Happiness
Studies. https://doi.org/10.1007/s10902-020-00298-6
Hello Stranger, 38
John, O. P., Donahue, E. M., & Kentle, R. L. (1991). The Big Five Inventory--Versions 4a and
54. Berkeley, CA: University of California, Berkeley, Institute of Personality and Social
Research.
Kardas, M., Kumar, A., & Epley, N. (2021). Overly shallow?: Miscalibrated expectations
create a barrier to deeper conversation. Journal of Personality and Social Psychology.
Kardas, M., Schroeder, J., & O’Brien, E. (2021). Keep talking: (Mis)understanding the hedonic trajectory
of conversation. Unpublished manuscript, Northwestern University.
Kushlev, K., Dwyer, R., & Dunn, E. W. (2019). The social price of constant connectivity:
Smartphones impose subtle costs on well-being. Current Directions in Psychological
Science, 28(4), 347-352. https://doi.org/10.1177/0963721419847200
https://doi.org/10.1177/0956797618772506
Lok, I., & Dunn, E. (in press). Can sociability be increased? J.P. Forgas, W. Crano & K. Fiedler
(Eds.), The Psychology of Sociability. New York: Routledge / Psychology Press.
Margolis, S., & Lyubomirsky, S. (2020). Experimental manipulation of extraverted and
introverted behavior and its effects on well-being. Journal of Experimental Psychology:
General, 149(4), 719–731. https://doi.org/10.1037/xge0000668
McNiel, J. M., & Fleeson, W. (2006). The causal effects of extraversion on positive affect and
neuroticism on negative affect: Manipulating state extraversion and state neuroticism in
an experimental approach. Journal of Research in Personality, 40, 529 –550.
McNiel, J. M., Lowman, J. C., & Fleeson, W. (2010). The effect of state extraversion on four
types of affect. European Journal of Personality, 24(1), 18-35.
Hello Stranger, 39
Murray, S. L., Holmes, J. G., & Collins, N. L. (2006). Optimizing assurance: The risk regulation
system in relationships. Psychological Bulletin, 132, 641-666.
https://doi.org/10.1037/0033-2909.132.5.641
Prentice, D., & Miller, D.T. (1996). Pluralistic ignorance and the perpetuation of social norms
by unwitting actors. Advances in Experimental Social Psychology, 28, 161-209.
https://doi.org/10.1016/S0065-2601(08)60238-5
Quoidbach, J., Taquet, M., Desseilles, M., de Montjoye, Y.-A., & Gross, J. J. (2019). Happiness
and Social Behavior. Psychological Science, 30(8), 1111–1122.
https://doi.org/10.1177/0956797619849666
Sandstrom, G., & Dunn, E. W. (2014). Is efficiency overrated? Minimal social interactions lead
to belonging and positive affect. Social Psychological & Personality Science, 5, 437–
442. https://doi.org/10.1177/1948550613502990
Sandstrom, G. M., & Boothby, E. J. (2020). Why do people avoid talking to strangers? A mini
meta-analysis of predicted fears and actual experiences talking to a stranger. Self and
Identity, 1-25. https://doi.org/10.1080/15298868.2020.1816568
Zelenski, J. M., Santoro, M. S., & Whelan, D. C. (2012). Would introverts be better off if they
acted more like extraverts? Exploring emotional and cognitive consequences of counter-
dispositional behavior. Emotion, 12, 290 –303. doi:10.1037/a0025169
Zelenski, J. M., Whelan, D. C., Nealis, L. J., Besner, C. M., Santoro, M. S., & Wynn, J. E.
(2013). Personality and affective forecasting: Trait introverts underpredict the hedonic
benefits of acting extraverted. Journal of Personality and Social Psychology, 104, 1092–
1108. doi:10.1037/a0032281
impact of compliments creates a barrier to expressing them. Journal of Personality and
Social Psychology.
Hello Stranger, 42
5.58% chose not to enroll (11.18% in the
conversation condition, 4.14% in the control
condition, 1.25% in the solitude condition)
Completed full survey (n = 383)
82.19% response rate (73.23% in the conversation
condition, 77.12% in the control condition, 84.06% in
the solitude condition)
condition, 100% in the control condition, 97.1% in
the solitude condition)
All conditions
(n = 383)
Talk to someone 14.4% 22.0% 6.8% 13.8%
Talk or socialize on computer or phone 34.7% 40.2% 32.2% 31.9%
Read 59.3% 58.3% 66.1% 54.3%
Sleep 26.9% 25.2% 30.5% 25.4%
Think 42.0% 39.4% 38.1% 47.8%
Work 36.8% 40.9% 37.3% 32.6%
Other 15.7% 11.0% 18.6% 17.4%
Today’s Commuting Behaviors
Talk to someone 28.7% 78.0% 5.9% 2.9%
Talk or socialize on computer or phone 26.6% 36.2% 30.5% 14.5%
Read 35.8% 29.9% 42.4% 35.5%
Sleep 12.5% 6.3% 16.1% 15.2%
Think 36.8% 27.6% 28.0% 52.9%
Work 15.7% 15.0% 19.5% 13.0%
Other 17.2% 12.6% 23.7% 15.9%
Extraversion 2.52 (0.83) 2.54 (0.88) 2.51 (0.78) 2.50 (0.84)
Commute duration 59.99 min
commute today? 70.0% yes 78.0% yes 66.9% yes 65.2% yes
Person with whom you spoke/Person sitting closest to you
Person’s Age - Older than you 44.6% 49.6% 45.8% 39.1%
Person’s Age - Younger than you 26.4% 22.0% 22.9% 33.3%
Person’s Age – Same as you 21.4% 23.6% 22.0% 18.8%
Hello Stranger, 45
Person’s Gender – Female 43.6% 52.0% 39.0% 39.9%
Person’s Gender – Male 53.3% 48.0% 57.6% 54.3%
Person’s Gender - Not sure 3.1% 0% 3.4% 5.8%
Person’s Race - Same race 71.3% 79.5% 72.0% 63.0%
Person’s Race - Different race 17.5% 14.2% 16.9% 21.0%
Person’s Race - Not sure 11.2% 6.3% 11.0% 15.9%
Conversation characteristics (only among those who talked)
Duration of conversation 10.28 min (10.88)
Interestingness of conversation
Impression of conversation-partner
Note. Standard deviations are reported in parentheses where appropriate. All measurements were taken after random assignment to
experimental condition, meaning that participants’ responses could have been influenced by their condition assignment. The normal
commuting behaviors and extraversion levels were reported in the pre-survey (prior to commute); all other variables were reported in
the post-survey (after commute). Ratings by experimental condition significantly varied on the following measures: normal talking
behavior, χ2 (2, 383) = 12.08, p = .002; all of the day’s reported commute behaviors except for reading and working; the gender of the
person by whom the participant sat or with whom the participant talked, χ2 (4, 383) = 14.57, p = .006; and the race of the person by
whom the participant sat or with whom the participant talked, χ2 (4, 383) = 10.03, p = .040.
Experiment 1: Instruction Cards
Experiment 1: Debriefing Statement
Thank you so much for taking part in the study earlier today.
The purpose of the study was to understand how different types of activities—specifically,
socializing compared to sitting in solitude—influence people’s well-being and happiness, and
people’s expectations about how these activities will influence their well-being.
We conducted a research study testing this question in 2014 using commuters in the Chicago
(USA) area. The study was published here: [paper link]. To read a few media articles describing
the study, you can see [link] or [link]. Overall, there were two key results from this paper. First,
commuters believed that conversing with a fellow commuter during their commute would be a
significantly more negative experience (e.g., less pleasant, provoking less happiness) than sitting
in solitude during the commute. Second, commuters’ beliefs were mistaken because talking to a
fellow commuter actually created a better experience (more pleasant, more happiness) than
sitting in solitude during the commute.
We have now partnered with the BBC World Service Group to conduct another test of this
finding in the UK. In today’s study, you were assigned to be in one of three experimental
conditions: talking to a fellow commuter, sitting in solitude, or doing what you would normally
do. We expect that being in the study today may have changed your commuting experience. We
will analyze your survey responses (in aggregate) in order to test our research question.
If you have further questions about today’s study, please contact [information].
Hello Stranger, 48
Predicted and Actual Positivity, Amount of Learning, and Productivity During Train Commutes
in the Conversation, Solitude, and Control Conditions in Experiment 1 (Intent to Treat; n = 382)
Positivity Learning Productivity
Pre-Commute
Post-Commute
Supplemental Table S3
Predicted and Actual Positivity, Amount of Learning, and Productivity During Train Commutes
in the Conversation, Solitude, and Control Conditions in Experiment 1 (Treated; n = 350)
Positivity Learning Productivity
Pre-Commute
Post-Commute
Hello Stranger, 49
To examine our primary hypothesis only among the treated participants (those who
reported following our instructions), we conducted a 2 (survey timing: pre-commute vs. post-
commute) × 3 (condition: conversation, solitude, or control) mixed model ANOVA on positivity.
There was a significant main effects of survey timing, F(1, 347) = 7.05, p = .008, ηp 2 = 0.02, and
condition, F(2, 347) = 20.53, p < .001, ηp 2 = 0.11, qualified by the predicted interaction, F(2,
347) = 4.27, p = .015, ηp 2 = 0.02 (Table S3). Consistent with our hypotheses, participants’ actual
experiences varied by experimental condition, F(2, 347) = 17.36, p < .001, ηp 2 = 0.09:
participants reported having a significantly more positive commute in the conversation condition
than in either the solitude condition, t(231) = 5.22, p < .001, or the control condition, t(215) =
4.72, p < .001. The difference between the solitude and control conditions was nonsignificant,
t(250) = -0.59. Inconsistent with our hypotheses, participants also expected the same overall
pattern of results across conditions before their commute, F(2, 347) = 14.90, p < .001, ηp 2 = 0.08,
expecting to have a more positive commute in the conversation condition than in the solitude
condition, t(231) = 3.42, p < .001, or the control condition, t(215) = 5.29, p < .001. They also
expected to have a more positive commute in the solitude than control condition, t(250) = 2.12, p
= .019.
Unsurprisingly, this “treated” analysis, which examines only those who followed our
experimental instructions and hence experienced the intended manipulation, are stronger than the
“intent-to-treat” analysis including all participants reported in the main text.
Hello Stranger, 50
Appendix A
Reasons Participants Provided For Not Talking in The Conversation Condition in Experiment 1.
• “He was wearing headphones.”
• “Everyone around me had headphones or working on their laptops.”
• “The train is a quiet place in the morning. There is an unwritten rule that you don't
communicate with others. Everyone around me had head phones on.”
• “The other commuters were reading and didn't want to disturb them.”
• “Anxiety and lack of opportunity. The latter because no one sat next to me, and the
people nearest me (across the aisle) were reading, on the phone, or had earphones in.
The former because I was about 20 years younger than everyone in my carriage and 1
of the only women. I felt uncomfortable striking up a conversation out of the blue.”
• “Too early in the morning.”
• “I needed to work.”
• “See: ‘how it made me feel’” [Note: in that section, the participant wrote: “I looked at
everyone on their phones with their earplugs in and felt awkward thinking about
striking up a conversation. So I chickened out. So now I feel frustrated and bit sad as I
see myself as a fairly gregarious person. Even though the train was delayed I couldn’t
think of a day to break the ice with anyone. I used to joke when I was backpacking
about striking up a conversation with random backpackers on a train, how I’d never do
it at home. Now I haven’t, again.”]
• “Too short a commute and too many folk engaged on devices and listening to music.”
• “No one sat close to me.”